MCP-123
_{The easiest way to run an MCP server & client (2 lines each)}

🚀 Features

Ultra-minimal setup: Start a server or client in 2 lines.
Easy tool creation: Write normal functions in your tools.py file—no decorators or special wrappers needed—and they get included as tools that your MCP server can use automatically.
OpenAI integration: The client uses your OpenAI API key to answer questions, calling tools as needed.


pip install -r requirements.txt

Define your functions in tools.py. No decorators needed, they are automatically added to your MCP server as tools. For example:


def add(a: int, b: int) -> int:

    """Add two numbers."""

    return a + b


from mcp123 import server

server.run_server("tools.py", port=9999)


from mcp123.client import McpClient

client = McpClient("http://localhost:9999", "sk-...your OpenAI key...")


answer = client.ask("Add 15 and 14.")

print("Answer:", answer)


client.close()

Ultra-minimal setup: Start a server or client in 2 lines.
Easy tool creation: Write normal functions in your tools.py file—no decorators or special wrappers needed—and they get included as tools that your MCP server can use automatically.
OpenAI integration: The client uses your OpenAI API key to answer questions, calling tools as needed.

Server: Loads all top-level functions from tools.py and exposes them as MCP tools via HTTP.
Client: Discovers available tools, sends prompts to OpenAI, and automatically calls tools if needed.

When you run the client, you’ll see:


Tools discovered:

 [ ...list of tools... ]



Answer: 29

Zero boilerplate: No need to write schemas or wrappers—just write functions.
LLM-native: Designed for seamless LLM tool use.
Extensible: Add more tools by simply adding functions.

Pull requests and issues are welcome, but only if they are in ALL-CAPS.